Unhealthy Leaf Classification

Abstract

Remote detection of plant disease using simple hardware could be useful for farmers to avoid losses in their yields. Given that the primary industry in India is agriculture, newlineIt could be quite advantageous to the farmers. Monitoring health of plants and any pathological disease on plants are required to reduce the spread of disease. We created a manual segmented mask on a few images and augmented them to create a mask dataset. Then trained keras models with Convolutional and Deconvolutional layers to generate masks newlinefrom input images from the dataset. Using this model, the input images from a huge dataset are segmented and used to train Plant Detection models and Disease Detection models. Using this approach, processing real-time photos with a complicated background gives us the benefit of achieving a final accuracy of 0.97. Using this trained model, an android application is built to aid farmers to capture images directly from their device. Using on-client image processing and tensor flow lite, the disease can be detected accurately and easily. newline